Generative models
Learning better visual representations and generation strategies, with a focus on diffusion models and image-to-video generation.
A LITTLE ORDER FROM NOISE PERSONAL / RESEARCH
Jinyuan Xu徐瑾源
Loading scene…
01
A little about me.
Hi, I’m Jinyuan Xu (徐瑾源). I work on making visual generative models better, faster, and more useful.
I’m a master’s student in Artificial Intelligence at The Hong Kong University of Science and Technology (Guangzhou), and a member of XLeaF Lab. My research focuses on video generation, diffusion models, and video restoration.
I’m also a research intern in the Imaging Algorithm Research Department at vivo.
Beyond research, I create AI-generated visual stories with 星火AI动漫. Across Kuaishou and Douyin, we have built a combined following of 10,000. Building with generative tools is also how I discover new research questions.
Efficient video restoration
Bringing high-quality video restoration closer to practical use through super-resolution, model distillation, and efficient sampling.
02
A few updates.
ON THE WAYJoined vivo as a research intern in the Imaging Algorithm Research Department.
Started my master’s studies in Artificial Intelligence at HKUST (Guangzhou).
Joined XLeaF Lab to explore visual generative models.
03
Selected research.
Better starting points. More efficient paths. Higher-quality videos.
View full diagram PDF ↗
Golden Anchor for
Video Generation
A good image is not always a good starting point for a video. Golden Anchor Learning trains an image generator with feedback from the videos it produces, learning anchors that better support downstream video generation.
Read research overview
Using a frozen image-to-video model and video-level rewards, the framework constructs cross-modal preference pairs and optimizes a text-to-image generator with offline DPO. Only a task LoRA is trained, adding no extra inference steps.
The project evaluates transfer across four image-to-video models, studying how learned anchors improve both visual quality and motion.
IMAGING ALGORITHM RESEARCH
Research in progress
Research in progress
Draft-and-Render: Mixed-Resolution Distillation for Video Super-Resolution
Sole first author / Ongoing · Targeting CVPR
EDUCATION
HKUST (Guangzhou)
Master’s · Artificial Intelligence
Anhui University
Bachelor’s · Statistics
RESEARCH EXPERIENCE
vivo
Research Intern · Imaging Algorithm Research
XLeaF Lab
Research Member · Visual Generative Models
Get in touch.
Happy to connect about generative models, video research,
or something interesting you’re building.
Alternate emailjxu957@connect.hkust-gz.edu.cn